[llvm] dc8d16c - [AArch64][GlobalISel] Add MMOs to constant pool loads to allow LICM hoisting.

[Amara Emerson via llvm-commits]

Author: Amara Emerson
Date: 2021-05-12T09:47:09-07:00
New Revision: dc8d16c03f4fcdc0dbfc2925621f6d0064bca31c

URL: https://github.com/llvm/llvm-project/commit/dc8d16c03f4fcdc0dbfc2925621f6d0064bca31c
DIFF: https://github.com/llvm/llvm-project/commit/dc8d16c03f4fcdc0dbfc2925621f6d0064bca31c.diff

LOG: [AArch64][GlobalISel] Add MMOs to constant pool loads to allow LICM hoisting.

This caused performance regressions vs SDAG on SingleSource/Benchmarks/Adobe-C++

Added: 
    llvm/test/CodeGen/AArch64/GlobalISel/select-const-pool.mir

Modified: 
    llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp b/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
index 0829f77151110..c763b7b3fae2a 100644
--- a/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
+++ b/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
@@ -31,6 +31,7 @@
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstr.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineMemOperand.h"
 #include "llvm/CodeGen/MachineOperand.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/TargetOpcodes.h"
@@ -3991,14 +3992,17 @@ AArch64InstructionSelector::emitConstantPoolEntry(const Constant *CPVal,
 
 MachineInstr *AArch64InstructionSelector::emitLoadFromConstantPool(
     const Constant *CPVal, MachineIRBuilder &MIRBuilder) const {
-  unsigned CPIdx = emitConstantPoolEntry(CPVal, MIRBuilder.getMF());
+  auto &MF = MIRBuilder.getMF();
+  unsigned CPIdx = emitConstantPoolEntry(CPVal, MF);
 
   auto Adrp =
       MIRBuilder.buildInstr(AArch64::ADRP, {&AArch64::GPR64RegClass}, {})
           .addConstantPoolIndex(CPIdx, 0, AArch64II::MO_PAGE);
 
   MachineInstr *LoadMI = nullptr;
-  switch (MIRBuilder.getDataLayout().getTypeStoreSize(CPVal->getType())) {
+  MachinePointerInfo PtrInfo = MachinePointerInfo::getConstantPool(MF);
+  unsigned Size = MIRBuilder.getDataLayout().getTypeStoreSize(CPVal->getType());
+  switch (Size) {
   case 16:
     LoadMI =
         &*MIRBuilder
@@ -4025,6 +4029,9 @@ MachineInstr *AArch64InstructionSelector::emitLoadFromConstantPool(
                       << *CPVal->getType());
     return nullptr;
   }
+  LoadMI->addMemOperand(MF, MF.getMachineMemOperand(PtrInfo,
+                                                    MachineMemOperand::MOLoad,
+                                                    Size, Align(Size)));
   constrainSelectedInstRegOperands(*Adrp, TII, TRI, RBI);
   constrainSelectedInstRegOperands(*LoadMI, TII, TRI, RBI);
   return LoadMI;

diff  --git a/llvm/test/CodeGen/AArch64/GlobalISel/select-const-pool.mir b/llvm/test/CodeGen/AArch64/GlobalISel/select-const-pool.mir
new file mode 100644
index 0000000000000..0e4e364e55714
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/select-const-pool.mir
@@ -0,0 +1,31 @@
+# RUN: llc -mtriple=aarch64-- -run-pass=instruction-select -verify-machineinstrs %s -o - | FileCheck %s
+---
+name:            test_constant_vec_pool_v2f64
+alignment:       4
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+liveins:
+  - { reg: '$x0' }
+frameInfo:
+  maxAlignment:    1
+machineFunctionInfo: {}
+body:             |
+  bb.1:
+    liveins: $x0
+    ; Check that we have an MMO on the load, which is needed for MachineLICM to hoist it.
+    ; CHECK-LABEL: name: test_constant_vec_pool_v2f64
+    ; CHECK: liveins: $x0
+    ; CHECK: [[COPY:%[0-9]+]]:gpr64sp = COPY $x0
+    ; CHECK: [[ADRP:%[0-9]+]]:gpr64common = ADRP target-flags(aarch64-page) %const.0
+    ; CHECK: [[LDRQui:%[0-9]+]]:fpr128 = LDRQui [[ADRP]], target-flags(aarch64-pageoff, aarch64-nc) %const.0 :: (load 16 from constant-pool)
+    ; CHECK: STRQui [[LDRQui]], [[COPY]], 0 :: (store 16)
+    ; CHECK: RET_ReallyLR
+    %0:gpr(p0) = COPY $x0
+    %3:fpr(s64) = G_FCONSTANT double 5.000000e-01
+    %2:fpr(s64) = G_FCONSTANT double 1.600000e+01
+    %1:fpr(<2 x s64>) = G_BUILD_VECTOR %2(s64), %3(s64)
+    G_STORE %1(<2 x s64>), %0(p0) :: (store 16)
+    RET_ReallyLR
+
+...